A LONGITUDINAL BRAIN-MACHINE INTERFACE TRAINING PARADIGM WITH A LOWER-LIMB EXOSKELETON & ITS INDUCED CORTICAL CHANGES

Introduction: Brain-machine interfaces (BMIs) have been developed to enable cognitive control of computers and robotic devices. Such technology might potentially lead to restoring movement for persons with motor disabilities by allowing them to control robotic prostheses or orthoses naturally with their mind. However, BMIs are still in their infancy, and long-term usage with closed-loop systems has not been thoroughly studied, nor the subsequent changes in the brain induced by cortical plasticity. Methods: Seven able-bodied subjects were recruited for a longitudinal BMI training paradigm with the Rex lower-limb exoskeleton. Participants developed their ability to use motor imagery over nine sessions to initiate the Rex’s walking and stopping as a Go-No Go task. The BMI consisted of active EEG processed through a Localized Fisher Discriminant Analysis dimensionality reduction and a Gaussian Mixture Model classifier on time-lagged δ band amplitudes. Training data were accumulated to update the decoding model over the first five sessions, after which model parameters were fixed for subjects to adapt to their personalized model. Subjects underwent a final session with simultaneous EEG-fMRI recording while watching video playback of themselves walking in the Rex performing the same motor imagery. Discussion: BMI decoding for control of the Rex’s gait varied among the subjects, with at least some achieving significantly above chance classification performance by the end of training. The fMRI scans showed contrasts in activation between the Walk and Stop conditions localized in the precentral gyrus among other areas associated with motor imagery. Offline EEG analysis identified ERPs corresponding to the walk cue, but these may not have been reliably detected by the classifier. Significance: The novelty in this study is the extended use of a subject pool continuously for many sessions of BMI training to control a walking exoskeleton. The longitudinal aspect provides insights into how much training subjects may need to achieve reliable classification, what factors separate good BMI operators from poor ones, and what other features may be more relevant in future BMI applications.